Biological mechanisms in vivo are extremely complicated cascades of signals, which are difficult to deconvolute and understand. Activation of T cells requires at least two signals. The recognition of the antigen by the T cell receptor is considered the first signal and the second signal arises from co-stimulation which results from the ligation of additional surface molecules on the T cell with additional molecules on an antigen presenting cell.
Thus T cell activation can be used to illustrate that the modulation of biological functions can require multiple signals. Other biological processes are equally complicated or more complicated. Whilst in vitro screening based on cells has and can assist with gaining insights into in vivo mechanisms the problem still arises of how to identify appropriate ligand pairs which modulate the biological function.
Bispecific antibodies are widely expected to play a major role in the next generation of biotherapeutics (D. Holmes, Nature Rev Drug Disc November 2011:10; 798). They have the potential to deliver superior, long term, broad efficacy in a greater proportion of patients. This can be achieved by either co-engaging different antigens simultaneously within a common disease pathway, thereby reducing redundancy; or by targeting antigens from independent pathways to provide an additive or synergistic effect.
Bispecific antibodies facilitate access to novel biology such as:                1) cross-linking receptors on a cell,        2) inducing cell mediated effects,        3) localizing a cytokine to a cell to regulate signaling or locally block cytokine function,        4) engaging multiple epitopes simultaneously to generate “new activity”, increase function or specificity, which may not be exhibited by a single monoclonal antibody or indeed mixtures of un-linked antibodies (‘poly-monoclonals’).        
Current strategies to engage dual targets are largely based on rational design of known mechanisms and include: cross-linking inhibitory receptors, co-engagement/clustering of receptors, blocking multiple stimulatory pathways, selective engagement of inhibitory receptors and blocking distinct pathways such as co-stimulation & cytokine signaling. However, the current state of the art in relation to known mechanisms and targets is a limiting factor to progress in this area.
Whilst bispecific antibodies have enormous potential as biological therapeutics they also present an increased set of challenges within discovery and development compared to monoclonal antibodies. Two key areas of difficulty are, 1) the development of a successful bispecific antibody format, and 2) selecting the pairs of targets to which the bispecific antibody will crosslink or co-engage.
Many promising bispecific antibody formats have now been developed that could potentially work as successful therapeutics including DVD-Ig (Abbvie), DuoBodies (Genmab), Knobs-in-Holes (Genentech), Common light chain (Merus). However, in each of these cases these formats are not ideally suited to high throughput target-dual-antigen discovery screening to enable the discovery of novel antigen pairs for crosslinking with bispecific antibodies. Typically for a single bispecific antibody construct at least two variable regions need to be sub-cloned from the original source of discovery vectors (e.g. phage display, hybridoma or single B-cell cloning) into appropriate bispecific expression vectors, each arm of the bispecific has to be expressed and the resulting bispecific antibody purified. This cloning and subsequent expression effort quickly becomes a significant practical bottleneck if large numbers of pairs of variable regions are to be combined in an attempt to screen for the most efficacious combination of discovered variable regions or to discover novel antigen pairs.
For example, if 50 unique antibodies are discovered against a panel of 50 cell surface targets, then a total of 2500 bispecific antibodies could potentially be generated (envisaged as an X-by-Y grid). With the bispecific antibody formats described above this would require at least 100 individual cloning reactions (50-X and 50-Y) followed by 2500 antibody expression experiments. Increasing the number of starting monoclonal antibodies to 100 would increase the minimal number of cloning reactions to 200 (100-X and 100-Y) and the expression number to 10,000.
Generally the root cause of this ‘expression bottleneck’ is the fact that the formats described above require both protein chain ‘halves’ of the final bispecific construct to be expressed simultaneously within a single expression experiment in the same cell. Therefore, for many formats, to produce 2500 bispecific antibodies, 2500 expression experiments are required.
The ‘expression bottleneck’ is further exacerbated if the bispecific antibody format is monocistronic (i.e. cloned and expressed as a single chain protein), for example single chain diabodies, as the number of cloning experiments would be 2500 and 10,000 respectively for the numbers given above.
Furthermore after expression, extensive purification may be required to isolate the desired construct.
Some bispecific approaches employ a common light chain in the bispecific constructs in order to reduce the amount of cloning, although this doesn't reduce the number of expression experiments. Furthermore, using a common chain, such as a common light chain, makes the challenge of antibody discovery harder as it is more difficult to find the starting antibody variable domains as the antibody needs to bind its antigen with a high enough affinity through one chain, such as the heavy chain, alone.
Accordingly the use of current bispecific formats in large scale and high throughput screening to identify novel antigen pairs is impractical and has led to the continued use of solely hypothesis driven approaches to bispecific antigen targeting.
We propose that rather than designing and testing a limited selection of bispecific antibodies that engage given epitopes on two known targets, the true potential of exploiting access to novel biology with bispecific antibodies can only be achieved through a broad functional screening effort with a large, diverse combinatorial panel of bispecific antibodies or protein ligands. To facilitate this screening a format and a method is required that enables the generation of large numbers of diverse bispecific proteins which can be readily constructed and screened for functional effects in a variety of functional screens. This approach allows for the serendipitous identification of synergistic pairs.
Thus it would be useful to generate and screen a large number of bispecific protein complexes present as combinations of various antigen specificities. In particular, it would be useful to be able to generate and screen a large number of different bispecific antibody complexes in a quick and efficient manner. There are a range of existing methods for manufacturing bispecific antibodies as already described above. However, each of these methods has its disadvantages, as do alternative methods as further described in more detail below.
The problem of how to efficiently identify targets for bispecific and multispecific constructs has not been adequately addressed in the art. For example, WO2014/001326 employs chemical conjugation of a protein to a DNA fragment, wherein the DNA fragment hybridises to a complementary DNA sequence that links two such proteins together for generating multispecific molecules comprising at least two targeting entities. There are number of difficulties associated with this approach if it were to be applied to identifying new bispecific combinations, for example conjugation of the protein to the DNA can result in damage to the activity and/or structure of the protein. In particular protein-DNA hybrids are not naturally occurring thus there is a potential for interference. In addition the chemical conjugation required for joining the protein and DNA adds complexity, time and expense to the process. Coupling and conjugation techniques exist for generating antibody drug conjugates and in vivo targeting technologies. Traditional chemical cross-linking is labour intensive as the relevant species may need to be purified from homodimers and other undesirable by-products. In addition, the chemical modification steps can alter the integrity of the proteins, thus leading to poor stability or altered biological function. As a result, the production of bispecific antibodies by chemical cross-linking is often inefficient and can also lead to a loss of antibody activity.
Another method of manufacturing bispecific antibodies is by cell-fusion (e.g. hybrid hybridomas), wherein the engineered cells express two heavy and two light antibody chains that assemble randomly. Since there are 4 possible variants to choose from, this results in the generation of 10 possible bispecific antibody combinations, of which only some (in many cases, only one) combinations would be desired. Hence, generating bispecific antibodies by cell-fusion results in low production yields and also requires an additional purification step in order to isolate the desired bispecific antibodies from the other bispecific antibodies produced. These disadvantages increase manufacturing time and costs.
Recombinant DNA techniques have also been employed for generating bispecific antibodies. For example, recombinant DNA techniques have also been used to generate ‘knob into hole’ bispecific antibodies. The ‘knob into hole’ technique involves engineering sterically complementary mutations in multimerization domains at the CH3 domain interface (see e.g., Ridgway et al., Protein Eng. 9:617-621 (1996); Merchant et al., Nat. Biotechnol. 16(7): 677-81 (1998); see also U.S. Pat. Nos. 5,731,168 and 7,183,076). One constraint of this strategy is that the light chains of the two parent antibodies have to be identical to prevent mispairing and formation of undesired and/or inactive molecules when expressed in the same cell. Each bispecific (heavy and light chains thereof) must be expressed in a single cell and the protein product generally contains about 20% of homodimer, which is subsequently removed by purification.
Other approaches are based on the natural exchange of chains in full-length IgG4 molecules (Genmab Duobody). However, this approach also has difficulties because it does not allow a construct to be prepared without an Fc region. As the Fc region can contribute to biological activity it may be difficult to establish if an activity observed is based on the combination of variable regions, the Fc or both in bispecific molecules comprising an Fc. Furthermore, the exchange is a dynamic process and this may lead to difficulties in relation to what the entity tested actually is.
Thus there is a need for new methods of generating bispecific protein complexes to enable the more efficient and higher throughput screening of bispecific antibodies. In particular, there is a need for a format and a method wherein a selection of any two antibodies or antibody fragments from a pool of available antibodies or antibody fragments can be readily combined to efficiently produce a multiplex of different bispecific antibodies, whilst, for example avoiding or minimising the formation of homodimers. Assembling different bispecific antibodies efficiently is particularly important when screening for synergistic biological function for new combinations of antigen specificities, in particular where heterodimers are essential for discovering that function.